Skip to content
All videos
newsDatabricks·July 7, 2025

Saving Millions From Millions: Navigating Towards Cost-Efficiency in Pinterest's Spark Jobs

Summary

Pinterest built a three-layer cost optimization system for Spark jobs using observability platforms (collecting real-time metrics and event logs), platform innovations (Apache Celeborn remote shuffle service and Kubernetes burst-aware memory allocation), and fine-grained management tools. Production deployments achieved 30-50% cost reductions per job through resource tuning while improving cluster utilization.

Summary generated by brickster.ai from the video transcript.

More from Databricks